Cyclic Barrier follow up #1417
Cyclic Barrier follow up #1417
Conversation
|
@TimWSpence FYI |
| case State(awaiting, epoch, unblock) => | ||
| val awaitingNow = awaiting - 1 | ||
|
|
||
| if (awaitingNow == 0) |
There was a problem hiding this comment.
nice, this invariant is much clearer
| s"$name - raise an exception when constructed with a negative capacity" in real { | ||
| val test = IO.defer(constructor(-1)).attempt | ||
| test.flatMap { res => | ||
| implicit class Assertions[A](fa: IO[A]) { |
There was a problem hiding this comment.
this would be super useful in BaseSpec
| case Left(e) => e must haveClass[TimeoutException] | ||
| }) | ||
| } yield res | ||
| s"$name - await releases all fibers" in real { |
There was a problem hiding this comment.
some tests use real and others use ticked, but i can't really tell why :)
There was a problem hiding this comment.
The ones that use ticked need to assert nontermination or cancelation, and ticked let's you do that much more easily. I used real whenever I could though
There was a problem hiding this comment.
nontermination makes sense, but how come cancellation is an issue?
s"$name - await is cancelable" in ticked { implicit ticker =>
for { newBarrier(2).flatMap(_.await).timeoutTo(1.second, IO.unit) must completeAs(())
i thought this was a good candidate for real but i'm probably missing something about how TC works
There was a problem hiding this comment.
oh, is it because cancellation is more or less nondeterministic on the sleep when using real?
There was a problem hiding this comment.
yeah, and ticked is deterministic and fast. With real you tradeoff slow test vs predictable test. Also these tests don't have any of the things that make ticked deadlock-prone (lots of fibers depending on each other) and untrustworthy (real memory barrier things, it's just Ref + Deferred + uncancelable)
Few changes:
awaitingNow == 0vsawaiting < capacity - 1